CN105005987A - SAR image superpixel generating method based on general gamma distribution - Google Patents
SAR image superpixel generating method based on general gamma distribution Download PDFInfo
- Publication number
- CN105005987A CN105005987A CN201510348973.9A CN201510348973A CN105005987A CN 105005987 A CN105005987 A CN 105005987A CN 201510348973 A CN201510348973 A CN 201510348973A CN 105005987 A CN105005987 A CN 105005987A
- Authority
- CN
- China
- Prior art keywords
- sar image
- pixel
- region
- image
- sar
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10032—Satellite or aerial image; Remote sensing
- G06T2207/10044—Radar image
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20021—Dividing image into blocks, subimages or windows
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20112—Image segmentation details
- G06T2207/20128—Atlas-based segmentation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30181—Earth observation
Landscapes
- Radar Systems Or Details Thereof (AREA)
Abstract
The invention provides a SAR image superpixel generating method based on general gamma distribution. The method comprises: dividing a SAR image into a plurality of areas and describing the PDF of each area of the SAR image by using gamma distribution, wherein the PDF represents probability density function; constructing a local clustering criterion by the combination of the PDF of the SAR image and spatial distance information, clustering each pixel of the SAR image, and updating the area divisions of the SAR image; and eliminating isolated small areas in the plurality of the areas of the SAR image by using a boundary evolution method based on a local Bayes criterion, thereby enabling each rest area to corresponds to a SAR image superpixel. The SAR image superpixel generating method may better utilize the context information of the SAR image, eliminates isolated small areas, and achieves more superposed generated superpixel boundary and a real object boundary.
Description
Technical field
The present invention relates to imaging radar technical field of image segmentation, more particularly, relate to a kind of SAR (Synthetic ApertureRadar, synthetic-aperture radar) image superpixel generation method.
Background technology
SAR is a kind of active microwave imaging sensor, has round-the-clock, round-the-clock imaging capability, and this makes SAR image be widely applied in a lot of military and civilian field.SAR image segmentation is the important content of SAR image automatic interpretation, is the important step that other SAR image decipher algorithms a large amount of comprise object detection and recognition, terrain classification etc.In recent years, along with the development of SAR imaging technique, the resolution of SAR image improves constantly.Compared with low resolution SAR image in traditional, High Resolution SAR Images can obtain the information of atural object and target more horn of plenty, how effectively to excavate the important research content that these effective informations are current SAR image interpretation.It is a kind of by carrying out cluster to image pixel that SAR image super-pixel generates, and obtains the method with the image-region of certain sense that a series of size is close.It is a kind of image partition method in essence that SAR image super-pixel generates, and rises, be with a wide range of applications in based on the SAR image decomposition method investigation and application system constructing in region along with the appearance of High Resolution SAR Images.
Current, typical image superpixel generation method has based on the Normalized-Cut method of graph theory, coagulation type cluster (agglomerative clustering) method, Turbopixel method and SLIC (Simple Linear Iterative Clustering, simple linear iteration cluster) method etc., its research is carried out mainly for optical imagery.Compared with additive method, SLIC method all has more excellent performance on edge fitting and efficiency.
SLIC method mainly comprises two steps: (1) utilizes local k-mean algorithm to image pixel cluster; (2) connected component algorithm is utilized to eliminate isolated image zonule.SLIC method, in cluster process, make use of each pixel and carries out cluster to the Euclidean distance of cluster centre (i.e. the average of image-region).But by the impact of the intrinsic coherent speckle noise of SAR image and the potential Larger Dynamic variation characteristic of SAR image range value, SLIC method is difficult to generate effective SAR image super-pixel.In addition, connected component algorithm eliminates isolated zonule by means of only minimum distance principle, does not make full use of the feature of image itself, is difficult to the problem that the border of the initial super-pixel that treatment step (1) generates and actual atural object border are departed from.Due to PDF (Probability DensityFunction, probability density function) statistical property in SAR image region can more fully be described than the average of image-region, and broad sense gamma distribution is a kind of important PDF for describing SAR image Region Statistical Features, therefore, the SAR image super-pixel generation method studied based on broad sense gamma distribution has important theory significance and using value.
Summary of the invention
The present invention, in order to effectively solve SAR image super-pixel Generating Problems, provides a kind of SAR image super-pixel generation method based on broad sense gamma distribution.This method can generate good super-pixel of coincideing with the border of actual atural object, can provide important support for the SAR image decomposition method investigation and application system constructing based on region.
Technical scheme of the present invention is: a kind of SAR image super-pixel generation method based on broad sense gamma distribution, is divided into some regions by SAR image, it is characterized in that, comprise the steps:
Utilize broad sense gamma to distribute and describe the region PDF of SAR image; In conjunction with region PDF and the space length information structuring Local Clustering criterion of SAR image, cluster is carried out to each pixel of SAR image, obtains some regions of SAR image;
Utilize based on the isolated zonule in the evolution method elimination some regions of SAR image, border of local bayesian criterion, the then corresponding SAR image super-pixel in remaining each region.
Especially, Local Clustering criterion judges based on pixel to be sorted and the amplitude similarity at center to be clustered and the weighted sum of Distance conformability degree.
The invention has the beneficial effects as follows: the statistical information that more fully can utilize region in conjunction with the region PDF of SAR image and space length information structuring Local Clustering criterion, generate the better initial pictures region of homogeneity.Utilize the border evolution method based on local bayesian criterion can utilize the contextual information of SAR image better, isolated zonule can not only be eliminated, the border of the border of the super-pixel generated and actual atural object can also be made more identical.
Accompanying drawing explanation
Fig. 1 is that SAR image super-pixel of the present invention generates method flow diagram;
Fig. 2 utilizes measuring image to carry out the result of testing, wherein (a) original SAR image; B () adopts the result figure of standard SLIC; C () utilizes result figure of the present invention;
Fig. 3 is the experimental result in region 1 in corresponding measuring image, wherein the region 1 of (a) original SAR image; B () adopts the result figure of standard SLIC; C () utilizes result figure of the present invention;
Fig. 4 is the experimental result in region 2 in corresponding measuring image, wherein the region 2 of (a) original SAR image; B () adopts the result figure of standard SLIC; C () utilizes result figure of the present invention;
Embodiment
Below in conjunction with accompanying drawing, SAR image super-pixel generation method provided by the invention is described in detail.
Fig. 1 is that SAR image super-pixel of the present invention generates method flow diagram.The first step of this flow process is algorithm initialization, comprises SAR image standardization, SAR image region initial division, the initialization of SAR image pixel tag, the initialization of SAR image cluster centre and similarity matrix initialization.Second step is the SAR image Local Clustering of region PDF based on SAR image and space length information.Each cluster centre corresponding, calculates each pixel in its hunting zone and the comprehensive similarity between it, utilizes it to upgrade similarity matrix, and then upgrades the label of respective pixel.3rd step is develop based on the SAR image zone boundary of local bayesian criterion.Owing to being subject to the impact of the factors such as coherent speckle noise, the SAR image region that second step obtains comprises a lot of isolated zonule, and certain deviation may be there is with the border of actual atural object in the border in SAR image region, process by adopting the border based on local bayesian criterion to develop, not only can eliminate isolated zonule, also make the border of the border of the super-pixel generated and atural object more identical.
Illustrate each step above-mentioned below:
The first step: algorithm initialization.
Step 1., SAR image standardization.If a given original SAR image I
0range value be I
0(x, y), 1≤x≤M, 1≤y≤N, wherein (x, y) represents the pixel coordinate value of this SAR image.To original SAR image I
0carry out standardization and can obtain standardization SAR image I
1,its range value is I
1(x, y)=I
0(x, y)/μ
0, 1≤x≤M, 1≤y≤N, wherein
for original SAR image I
0average.Aims of standardization are the dynamic ranges in order to compressed image, are convenient to construct rational pixel clustering criteria.
Step 2., SAR image region initial division.By standardization SAR image I
1be divided into a series of adjacent size and be L
s× L
sthe image-region R of pixel
1, R
2..., R
k..., R
ns, namely
wherein
for region sum, be convenience of calculation, suppose that M and N is L
sintegral multiple, and L
svalue determine as required.
Step 3., the initialization of SAR image pixel tag.Give a label identical with image-region sequence number to the pixel 2. divided by step in each image-region of obtaining, obtain label image L, that is:
Step 4., the initialization of SAR image cluster centre.Normalized SAR image I
1gradient-norm image G, its each pixel G (x, y) is calculated by following formula:
Then by image-region R
kin there is the pixel of minimal gradient mould as image-region R
kinitial cluster center C
kif, that is:
Then make C
k=I
1(x
k, y
k).
Step 5., similarity matrix initialization.Initialization similarity matrix S is the full null matrix that the capable N of M arranges, even S={S (x, y) | S (x, y)=0,1≤x≤M, 1≤y≤N}.Similarity matrix record standard SAR image I
1in each pixel and the similarity of certain cluster centre.
Second step: based on the PDF of image-region and the SAR image Local Clustering of space length information.
1. step, estimates standardization SAR image I
1in the PDF of each image-region.If the PDF in each region of image is broad sense gamma distribution, a kth image-region R
kthe form parameter of PDF, energy parameter and scale parameter be
calculate in the following manner:
In above formula:
Wherein:
For sign function, Ψ (1) and Ψ () is respectively the 1st rank polygamma function and digamma function, N
rkrepresent image-region R
kin number of pixels.
Step 2., based on the PDF of image-region and the Local Clustering of space length information.
A the hunting zone of () cluster centre is determined.Order is with C
kcentered by size be 2L
s+ 1 × 2L
sthe window ranges of+1 is cluster centre C
khunting zone, k=1,2 ..., N
s.
B () comprehensive similarity calculates.Corresponding each cluster centre C respectively
k, calculate C successively
kand the comprehensive similarity in its hunting zone between each pixel
for:
Wherein:
represent at cluster centre C
khunting zone in m pixel at standardization SAR image I
1in range value,
represent C
knumber of pixels in hunting zone;
C () similarity matrix S, SAR image pixel tag and image-region upgrade.
If
Then make
By image I
1the pixel of middle label L (x, y)=k incorporates image-region R into
kin, upgrade simultaneously
3., cluster centre upgrades step.
4., iteration performs step 1. to 3. to step, until cluster centre C
kcoordinate figure no longer to change or iteration number of times reaches the maximal value of setting
3rd step: the SAR image zone boundary based on local bayesian criterion develops.
1., boundary pixel maximum likelihood develops step.Make standardization SAR image I
1image-region R
1, R
2..., R
k...,
boundary pixel set be
wherein N
bfor the sum of boundary pixel, boundary pixel
coordinate figure be
then according to following maximum-likelihood criterion to each boundary pixel
label upgrade:
Wherein
for boundary pixel
second order neighborhood in label matrix, that is:
2., 1. iteration performs step to step, until iterations reaches the maximum times of setting
3., boundary pixel maximum a posteriori probability develops step.According to following maximum posteriori criterion to each boundary pixel
label again upgrade:
Wherein:
For weight matrix, smoothing parameter α > 0, typically desirable α=1.5;
for
oriental matrix:
For
the i-th ' row jth ' column element.
Step 4., repeated execution of steps 3., until perform number of times to reach the maximum times of setting
then by original image I
0the pixel of middle label L (x, y)=k incorporates image-region R into
k,k=1,2 ..., N
s,then image-region R
kbe original image I
0a kth super-pixel.
Fig. 2 (a) gives the original SAR image that a width size is 300 × 450 pixels, Fig. 2 (b) and Fig. 2 (c) sets forth the SAR image super-pixel result of employing standard SLIC and the inventive method generation, and wherein in the inventive method, optimum configurations is L
s=15, w=0.6,
with
α=1.5.As can be seen from the figure, the border of super-pixel that standard SLIC method generates can better coincide the border of the larger atural object of contrast, but coincide poor for the border of the less atural object of contrast.By contrast, the border of the atural object that the border of super-pixel that the inventive method generates is not only large with contrast coincide very well, and the border of also relatively little with contrast atural object coincide better.
In order to the SAR image super-pixel result that observation caliber SLIC method better and the inventive method generate, Fig. 3 and Fig. 4 sets forth and utilize two kinds of methods to the result of representative region (namely respectively the region 1 shown in the rectangle frame of the lower left corner and the region 2 shown in the rectangle frame of the lower right corner in corresponding diagram 2 (a)) comprising different atural object.Fig. 3 (a) gives the original SAR image corresponding with region 1, Fig. 3 (b) and Fig. 3 (c) sets forth the super-pixel result corresponding with region 1 of employing standard SLIC and the inventive method generation, and wherein curve represents the border of super-pixel; Fig. 4 (a) gives the original SAR image corresponding with region 2, Fig. 4 (b) and Fig. 4 (c) sets forth the original SAR image super-pixel result corresponding with region 2 of employing standard SLIC and the inventive method generation, and wherein curve represents the border of super-pixel.
Can clearly be seen that from Fig. 3 and Fig. 4, compared with standard SLIC method, the inventive method can generate the super-pixel of more coincideing with actual atural object border, in region B and Fig. 4 (a) shown in region A typically as shown in left side central portion ellipse in corresponding diagram 3 (a) and lower right corner ellipse, in the middle of bottom, the super-pixel of the region C shown in ellipse generates result, thus demonstrates the validity of the inventive method.
Claims (2)
1. the SAR image super-pixel generation method based on broad sense gamma distribution, SAR refers to synthetic-aperture radar, SAR image is divided into some regions, it is characterized in that, also comprise the steps:
Utilize broad sense gamma to distribute and describe the region PDF in each region of SAR image, PDF refers to probability density function; In conjunction with region PDF and the space length information structuring Local Clustering criterion of SAR image, cluster is carried out to each pixel of SAR image, upgrade the Region dividing of SAR image; Utilize based on the isolated zonule in the evolution method elimination some regions of SAR image, border of local bayesian criterion, the then corresponding SAR image super-pixel in remaining each region.
2. the SAR image super-pixel generation method based on broad sense gamma distribution according to claim 1, is characterized in that, Local Clustering criterion judges based on pixel to be sorted and the amplitude similarity at center to be clustered and the weighted sum of Distance conformability degree.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201510348973.9A CN105005987A (en) | 2015-06-23 | 2015-06-23 | SAR image superpixel generating method based on general gamma distribution |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201510348973.9A CN105005987A (en) | 2015-06-23 | 2015-06-23 | SAR image superpixel generating method based on general gamma distribution |
Publications (1)
Publication Number | Publication Date |
---|---|
CN105005987A true CN105005987A (en) | 2015-10-28 |
Family
ID=54378644
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201510348973.9A Pending CN105005987A (en) | 2015-06-23 | 2015-06-23 | SAR image superpixel generating method based on general gamma distribution |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN105005987A (en) |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105426914A (en) * | 2015-11-19 | 2016-03-23 | 中国人民解放军信息工程大学 | Image similarity detection method for position recognition |
CN105549009A (en) * | 2015-12-09 | 2016-05-04 | 电子科技大学 | SAR image CFAR target detection method based on super pixels |
CN106778821A (en) * | 2016-11-25 | 2017-05-31 | 西安电子科技大学 | Classification of Polarimetric SAR Image method based on SLIC and improved CNN |
CN107016684A (en) * | 2017-04-13 | 2017-08-04 | 中国人民解放军国防科学技术大学 | A kind of super-pixel rapid generation of Polarimetric SAR Image |
CN110717956A (en) * | 2019-09-30 | 2020-01-21 | 重庆大学 | L0 norm optimization reconstruction method guided by finite angle projection superpixel |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20130188869A1 (en) * | 2012-01-20 | 2013-07-25 | Korea Advanced Institute Of Science And Technology | Image segmentation method using higher-order clustering, system for processing the same and recording medium for storing the same |
WO2014166709A1 (en) * | 2013-04-12 | 2014-10-16 | Thomson Licensing | Superpixel generation with improved spatial coherency |
CN104361351A (en) * | 2014-11-12 | 2015-02-18 | 中国人民解放军国防科学技术大学 | Synthetic aperture radar (SAR) image classification method on basis of range statistics similarity |
CN104463210A (en) * | 2014-12-08 | 2015-03-25 | 西安电子科技大学 | Polarization SAR image classification method based on object orienting and spectral clustering |
WO2015049118A1 (en) * | 2013-10-02 | 2015-04-09 | Thomson Licensing | Method and apparatus for generating superpixel clusters |
CN104680538A (en) * | 2015-03-09 | 2015-06-03 | 西安电子科技大学 | SAR image CFAR target detection method on basis of super pixels |
-
2015
- 2015-06-23 CN CN201510348973.9A patent/CN105005987A/en active Pending
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20130188869A1 (en) * | 2012-01-20 | 2013-07-25 | Korea Advanced Institute Of Science And Technology | Image segmentation method using higher-order clustering, system for processing the same and recording medium for storing the same |
WO2014166709A1 (en) * | 2013-04-12 | 2014-10-16 | Thomson Licensing | Superpixel generation with improved spatial coherency |
WO2015049118A1 (en) * | 2013-10-02 | 2015-04-09 | Thomson Licensing | Method and apparatus for generating superpixel clusters |
CN104361351A (en) * | 2014-11-12 | 2015-02-18 | 中国人民解放军国防科学技术大学 | Synthetic aperture radar (SAR) image classification method on basis of range statistics similarity |
CN104463210A (en) * | 2014-12-08 | 2015-03-25 | 西安电子科技大学 | Polarization SAR image classification method based on object orienting and spectral clustering |
CN104680538A (en) * | 2015-03-09 | 2015-06-03 | 西安电子科技大学 | SAR image CFAR target detection method on basis of super pixels |
Non-Patent Citations (3)
Title |
---|
FACHAO QIN 等: "Superpixel Segmentation for Polarimetric SAR Imagery Using Local Iterative Clustering", 《IEEE GEOSCIENCE AND REMOTE SENSING LETTERS》 * |
XIANXIANG QIN 等: "SAR Image Segmentation via Hierarchical Region Merging and Edge Evolving With Generalized Gamma Distribution", 《IEEE GEOSCIENCE AND REMOTE SENSING LETTERS》 * |
张亚亚 等: "基于改进SLIC方法的彩色图像分割", 《计算机工程》 * |
Cited By (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105426914A (en) * | 2015-11-19 | 2016-03-23 | 中国人民解放军信息工程大学 | Image similarity detection method for position recognition |
CN105426914B (en) * | 2015-11-19 | 2019-03-15 | 中国人民解放军信息工程大学 | A kind of image similarity detection method of facing position identification |
CN105549009A (en) * | 2015-12-09 | 2016-05-04 | 电子科技大学 | SAR image CFAR target detection method based on super pixels |
CN105549009B (en) * | 2015-12-09 | 2017-12-15 | 电子科技大学 | A kind of SAR image CFAR object detection methods based on super-pixel |
CN106778821A (en) * | 2016-11-25 | 2017-05-31 | 西安电子科技大学 | Classification of Polarimetric SAR Image method based on SLIC and improved CNN |
CN107016684A (en) * | 2017-04-13 | 2017-08-04 | 中国人民解放军国防科学技术大学 | A kind of super-pixel rapid generation of Polarimetric SAR Image |
CN110717956A (en) * | 2019-09-30 | 2020-01-21 | 重庆大学 | L0 norm optimization reconstruction method guided by finite angle projection superpixel |
CN110717956B (en) * | 2019-09-30 | 2023-06-20 | 重庆大学 | L0 norm optimization reconstruction method guided by limited angle projection superpixel |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Han et al. | A deep learning method for bias correction of ECMWF 24–240 h forecasts | |
CN108038445B (en) | SAR automatic target identification method based on multi-view deep learning framework | |
Solomon et al. | Wasserstein propagation for semi-supervised learning | |
CN101551856B (en) | SAR target recognition method based on sparse least squares support vector machine | |
CN105005987A (en) | SAR image superpixel generating method based on general gamma distribution | |
Yu et al. | A lightweight complex-valued DeepLabv3+ for semantic segmentation of PolSAR image | |
CN103955701B (en) | Multi-level-combined multi-look synthetic aperture radar image target recognition method | |
CN103258324B (en) | Based on the method for detecting change of remote sensing image that controlled kernel regression and super-pixel are split | |
US9449395B2 (en) | Methods and systems for image matting and foreground estimation based on hierarchical graphs | |
CN103810704A (en) | SAR (synthetic aperture radar) image change detection method based on support vector machine and discriminative random field | |
CN102402685A (en) | Method for segmenting three Markov field SAR image based on Gabor characteristic | |
Wang et al. | Dynamic graph-level neural network for SAR image change detection | |
Park et al. | Bayesian manifold learning: the locally linear latent variable model (LL-LVM) | |
CN103366371A (en) | K distribution and texture feature-based SAR (Synthetic Aperture Radar) image segmentation method | |
CN109146925A (en) | Conspicuousness object detection method under a kind of dynamic scene | |
Liu et al. | Target recognition in synthetic aperture radar images via joint multifeature decision fusion | |
Wu et al. | Cost-sensitive latent space learning for imbalanced PolSAR image classification | |
Yang et al. | CEMA-LSTM: Enhancing contextual feature correlation for radar extrapolation using fine-grained echo datasets | |
CN114004998B (en) | Unsupervised polarization SAR image ground object classification method based on multi-vision tensor product diffusion | |
Leinonen et al. | Unsupervised classification of snowflake images using a generative adversarial network and K-medoids classification | |
CN104123563B (en) | Cloude characteristic decomposition based polarimetric SAR (Synthetic Aperture Radar) image non-supervision classification method | |
CN103871060B (en) | Image partition method based on steady direction wave zone probability graph model | |
Giroux et al. | T-fftradnet: Object detection with swin vision transformers from raw adc radar signals | |
Liu et al. | Pseudo-label growth dictionary pair learning for crowd counting | |
CN110110670A (en) | Data correlation method in pedestrian tracking based on Wasserstein measurement |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
RJ01 | Rejection of invention patent application after publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20151028 |